| 1 | /* |
| 2 | * TCP over IPv6 |
| 3 | * Linux INET6 implementation |
| 4 | * |
| 5 | * Authors: |
| 6 | * Pedro Roque <roque@di.fc.ul.pt> |
| 7 | * |
| 8 | * Based on: |
| 9 | * linux/net/ipv4/tcp.c |
| 10 | * linux/net/ipv4/tcp_input.c |
| 11 | * linux/net/ipv4/tcp_output.c |
| 12 | * |
| 13 | * Fixes: |
| 14 | * Hideaki YOSHIFUJI : sin6_scope_id support |
| 15 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which |
| 16 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind |
| 17 | * a single port at the same time. |
| 18 | * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file. |
| 19 | * |
| 20 | * This program is free software; you can redistribute it and/or |
| 21 | * modify it under the terms of the GNU General Public License |
| 22 | * as published by the Free Software Foundation; either version |
| 23 | * 2 of the License, or (at your option) any later version. |
| 24 | */ |
| 25 | |
| 26 | #include <linux/bottom_half.h> |
| 27 | #include <linux/module.h> |
| 28 | #include <linux/errno.h> |
| 29 | #include <linux/types.h> |
| 30 | #include <linux/socket.h> |
| 31 | #include <linux/sockios.h> |
| 32 | #include <linux/net.h> |
| 33 | #include <linux/jiffies.h> |
| 34 | #include <linux/in.h> |
| 35 | #include <linux/in6.h> |
| 36 | #include <linux/netdevice.h> |
| 37 | #include <linux/init.h> |
| 38 | #include <linux/jhash.h> |
| 39 | #include <linux/ipsec.h> |
| 40 | #include <linux/times.h> |
| 41 | #include <linux/slab.h> |
| 42 | #include <linux/uaccess.h> |
| 43 | #include <linux/ipv6.h> |
| 44 | #include <linux/icmpv6.h> |
| 45 | #include <linux/random.h> |
| 46 | |
| 47 | #include <net/tcp.h> |
| 48 | #include <net/ndisc.h> |
| 49 | #include <net/inet6_hashtables.h> |
| 50 | #include <net/inet6_connection_sock.h> |
| 51 | #include <net/ipv6.h> |
| 52 | #include <net/transp_v6.h> |
| 53 | #include <net/addrconf.h> |
| 54 | #include <net/ip6_route.h> |
| 55 | #include <net/ip6_checksum.h> |
| 56 | #include <net/inet_ecn.h> |
| 57 | #include <net/protocol.h> |
| 58 | #include <net/xfrm.h> |
| 59 | #include <net/snmp.h> |
| 60 | #include <net/dsfield.h> |
| 61 | #include <net/timewait_sock.h> |
| 62 | #include <net/inet_common.h> |
| 63 | #include <net/secure_seq.h> |
| 64 | #include <net/tcp_memcontrol.h> |
| 65 | #include <net/busy_poll.h> |
| 66 | |
| 67 | #include <linux/proc_fs.h> |
| 68 | #include <linux/seq_file.h> |
| 69 | |
| 70 | #include <linux/crypto.h> |
| 71 | #include <linux/scatterlist.h> |
| 72 | |
| 73 | static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb); |
| 74 | static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, |
| 75 | struct request_sock *req); |
| 76 | |
| 77 | static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb); |
| 78 | |
| 79 | static const struct inet_connection_sock_af_ops ipv6_mapped; |
| 80 | static const struct inet_connection_sock_af_ops ipv6_specific; |
| 81 | #ifdef CONFIG_TCP_MD5SIG |
| 82 | static const struct tcp_sock_af_ops tcp_sock_ipv6_specific; |
| 83 | static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific; |
| 84 | #else |
| 85 | static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, |
| 86 | const struct in6_addr *addr) |
| 87 | { |
| 88 | return NULL; |
| 89 | } |
| 90 | #endif |
| 91 | |
| 92 | static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) |
| 93 | { |
| 94 | struct dst_entry *dst = skb_dst(skb); |
| 95 | |
| 96 | if (dst) { |
| 97 | const struct rt6_info *rt = (const struct rt6_info *)dst; |
| 98 | |
| 99 | dst_hold(dst); |
| 100 | sk->sk_rx_dst = dst; |
| 101 | inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; |
| 102 | inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt); |
| 103 | } |
| 104 | } |
| 105 | |
| 106 | static __u32 tcp_v6_init_sequence(const struct sk_buff *skb) |
| 107 | { |
| 108 | return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32, |
| 109 | ipv6_hdr(skb)->saddr.s6_addr32, |
| 110 | tcp_hdr(skb)->dest, |
| 111 | tcp_hdr(skb)->source); |
| 112 | } |
| 113 | |
| 114 | static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr, |
| 115 | int addr_len) |
| 116 | { |
| 117 | struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr; |
| 118 | struct inet_sock *inet = inet_sk(sk); |
| 119 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 120 | struct ipv6_pinfo *np = inet6_sk(sk); |
| 121 | struct tcp_sock *tp = tcp_sk(sk); |
| 122 | struct in6_addr *saddr = NULL, *final_p, final; |
| 123 | struct flowi6 fl6; |
| 124 | struct dst_entry *dst; |
| 125 | int addr_type; |
| 126 | int err; |
| 127 | |
| 128 | if (addr_len < SIN6_LEN_RFC2133) |
| 129 | return -EINVAL; |
| 130 | |
| 131 | if (usin->sin6_family != AF_INET6) |
| 132 | return -EAFNOSUPPORT; |
| 133 | |
| 134 | memset(&fl6, 0, sizeof(fl6)); |
| 135 | |
| 136 | if (np->sndflow) { |
| 137 | fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK; |
| 138 | IP6_ECN_flow_init(fl6.flowlabel); |
| 139 | if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { |
| 140 | struct ip6_flowlabel *flowlabel; |
| 141 | flowlabel = fl6_sock_lookup(sk, fl6.flowlabel); |
| 142 | if (!flowlabel) |
| 143 | return -EINVAL; |
| 144 | fl6_sock_release(flowlabel); |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | /* |
| 149 | * connect() to INADDR_ANY means loopback (BSD'ism). |
| 150 | */ |
| 151 | |
| 152 | if (ipv6_addr_any(&usin->sin6_addr)) |
| 153 | usin->sin6_addr.s6_addr[15] = 0x1; |
| 154 | |
| 155 | addr_type = ipv6_addr_type(&usin->sin6_addr); |
| 156 | |
| 157 | if (addr_type & IPV6_ADDR_MULTICAST) |
| 158 | return -ENETUNREACH; |
| 159 | |
| 160 | if (addr_type&IPV6_ADDR_LINKLOCAL) { |
| 161 | if (addr_len >= sizeof(struct sockaddr_in6) && |
| 162 | usin->sin6_scope_id) { |
| 163 | /* If interface is set while binding, indices |
| 164 | * must coincide. |
| 165 | */ |
| 166 | if (sk->sk_bound_dev_if && |
| 167 | sk->sk_bound_dev_if != usin->sin6_scope_id) |
| 168 | return -EINVAL; |
| 169 | |
| 170 | sk->sk_bound_dev_if = usin->sin6_scope_id; |
| 171 | } |
| 172 | |
| 173 | /* Connect to link-local address requires an interface */ |
| 174 | if (!sk->sk_bound_dev_if) |
| 175 | return -EINVAL; |
| 176 | } |
| 177 | |
| 178 | if (tp->rx_opt.ts_recent_stamp && |
| 179 | !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) { |
| 180 | tp->rx_opt.ts_recent = 0; |
| 181 | tp->rx_opt.ts_recent_stamp = 0; |
| 182 | tp->write_seq = 0; |
| 183 | } |
| 184 | |
| 185 | sk->sk_v6_daddr = usin->sin6_addr; |
| 186 | np->flow_label = fl6.flowlabel; |
| 187 | |
| 188 | /* |
| 189 | * TCP over IPv4 |
| 190 | */ |
| 191 | |
| 192 | if (addr_type == IPV6_ADDR_MAPPED) { |
| 193 | u32 exthdrlen = icsk->icsk_ext_hdr_len; |
| 194 | struct sockaddr_in sin; |
| 195 | |
| 196 | SOCK_DEBUG(sk, "connect: ipv4 mapped\n"); |
| 197 | |
| 198 | if (__ipv6_only_sock(sk)) |
| 199 | return -ENETUNREACH; |
| 200 | |
| 201 | sin.sin_family = AF_INET; |
| 202 | sin.sin_port = usin->sin6_port; |
| 203 | sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; |
| 204 | |
| 205 | icsk->icsk_af_ops = &ipv6_mapped; |
| 206 | sk->sk_backlog_rcv = tcp_v4_do_rcv; |
| 207 | #ifdef CONFIG_TCP_MD5SIG |
| 208 | tp->af_specific = &tcp_sock_ipv6_mapped_specific; |
| 209 | #endif |
| 210 | |
| 211 | err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin)); |
| 212 | |
| 213 | if (err) { |
| 214 | icsk->icsk_ext_hdr_len = exthdrlen; |
| 215 | icsk->icsk_af_ops = &ipv6_specific; |
| 216 | sk->sk_backlog_rcv = tcp_v6_do_rcv; |
| 217 | #ifdef CONFIG_TCP_MD5SIG |
| 218 | tp->af_specific = &tcp_sock_ipv6_specific; |
| 219 | #endif |
| 220 | goto failure; |
| 221 | } |
| 222 | np->saddr = sk->sk_v6_rcv_saddr; |
| 223 | |
| 224 | return err; |
| 225 | } |
| 226 | |
| 227 | if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr)) |
| 228 | saddr = &sk->sk_v6_rcv_saddr; |
| 229 | |
| 230 | fl6.flowi6_proto = IPPROTO_TCP; |
| 231 | fl6.daddr = sk->sk_v6_daddr; |
| 232 | fl6.saddr = saddr ? *saddr : np->saddr; |
| 233 | fl6.flowi6_oif = sk->sk_bound_dev_if; |
| 234 | fl6.flowi6_mark = sk->sk_mark; |
| 235 | fl6.fl6_dport = usin->sin6_port; |
| 236 | fl6.fl6_sport = inet->inet_sport; |
| 237 | |
| 238 | final_p = fl6_update_dst(&fl6, np->opt, &final); |
| 239 | |
| 240 | security_sk_classify_flow(sk, flowi6_to_flowi(&fl6)); |
| 241 | |
| 242 | dst = ip6_dst_lookup_flow(sk, &fl6, final_p); |
| 243 | if (IS_ERR(dst)) { |
| 244 | err = PTR_ERR(dst); |
| 245 | goto failure; |
| 246 | } |
| 247 | |
| 248 | if (!saddr) { |
| 249 | saddr = &fl6.saddr; |
| 250 | sk->sk_v6_rcv_saddr = *saddr; |
| 251 | } |
| 252 | |
| 253 | /* set the source address */ |
| 254 | np->saddr = *saddr; |
| 255 | inet->inet_rcv_saddr = LOOPBACK4_IPV6; |
| 256 | |
| 257 | sk->sk_gso_type = SKB_GSO_TCPV6; |
| 258 | __ip6_dst_store(sk, dst, NULL, NULL); |
| 259 | |
| 260 | if (tcp_death_row.sysctl_tw_recycle && |
| 261 | !tp->rx_opt.ts_recent_stamp && |
| 262 | ipv6_addr_equal(&fl6.daddr, &sk->sk_v6_daddr)) |
| 263 | tcp_fetch_timewait_stamp(sk, dst); |
| 264 | |
| 265 | icsk->icsk_ext_hdr_len = 0; |
| 266 | if (np->opt) |
| 267 | icsk->icsk_ext_hdr_len = (np->opt->opt_flen + |
| 268 | np->opt->opt_nflen); |
| 269 | |
| 270 | tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); |
| 271 | |
| 272 | inet->inet_dport = usin->sin6_port; |
| 273 | |
| 274 | tcp_set_state(sk, TCP_SYN_SENT); |
| 275 | err = inet6_hash_connect(&tcp_death_row, sk); |
| 276 | if (err) |
| 277 | goto late_failure; |
| 278 | |
| 279 | sk_set_txhash(sk); |
| 280 | |
| 281 | if (!tp->write_seq && likely(!tp->repair)) |
| 282 | tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32, |
| 283 | sk->sk_v6_daddr.s6_addr32, |
| 284 | inet->inet_sport, |
| 285 | inet->inet_dport); |
| 286 | |
| 287 | err = tcp_connect(sk); |
| 288 | if (err) |
| 289 | goto late_failure; |
| 290 | |
| 291 | return 0; |
| 292 | |
| 293 | late_failure: |
| 294 | tcp_set_state(sk, TCP_CLOSE); |
| 295 | __sk_dst_reset(sk); |
| 296 | failure: |
| 297 | inet->inet_dport = 0; |
| 298 | sk->sk_route_caps = 0; |
| 299 | return err; |
| 300 | } |
| 301 | |
| 302 | static void tcp_v6_mtu_reduced(struct sock *sk) |
| 303 | { |
| 304 | struct dst_entry *dst; |
| 305 | |
| 306 | if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) |
| 307 | return; |
| 308 | |
| 309 | dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info); |
| 310 | if (!dst) |
| 311 | return; |
| 312 | |
| 313 | if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { |
| 314 | tcp_sync_mss(sk, dst_mtu(dst)); |
| 315 | tcp_simple_retransmit(sk); |
| 316 | } |
| 317 | } |
| 318 | |
| 319 | static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, |
| 320 | u8 type, u8 code, int offset, __be32 info) |
| 321 | { |
| 322 | const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; |
| 323 | const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); |
| 324 | struct net *net = dev_net(skb->dev); |
| 325 | struct request_sock *fastopen; |
| 326 | struct ipv6_pinfo *np; |
| 327 | struct tcp_sock *tp; |
| 328 | __u32 seq, snd_una; |
| 329 | struct sock *sk; |
| 330 | int err; |
| 331 | |
| 332 | sk = __inet6_lookup_established(net, &tcp_hashinfo, |
| 333 | &hdr->daddr, th->dest, |
| 334 | &hdr->saddr, ntohs(th->source), |
| 335 | skb->dev->ifindex); |
| 336 | |
| 337 | if (!sk) { |
| 338 | ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev), |
| 339 | ICMP6_MIB_INERRORS); |
| 340 | return; |
| 341 | } |
| 342 | |
| 343 | if (sk->sk_state == TCP_TIME_WAIT) { |
| 344 | inet_twsk_put(inet_twsk(sk)); |
| 345 | return; |
| 346 | } |
| 347 | seq = ntohl(th->seq); |
| 348 | if (sk->sk_state == TCP_NEW_SYN_RECV) |
| 349 | return tcp_req_err(sk, seq); |
| 350 | |
| 351 | bh_lock_sock(sk); |
| 352 | if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG) |
| 353 | NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); |
| 354 | |
| 355 | if (sk->sk_state == TCP_CLOSE) |
| 356 | goto out; |
| 357 | |
| 358 | if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) { |
| 359 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); |
| 360 | goto out; |
| 361 | } |
| 362 | |
| 363 | tp = tcp_sk(sk); |
| 364 | /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ |
| 365 | fastopen = tp->fastopen_rsk; |
| 366 | snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; |
| 367 | if (sk->sk_state != TCP_LISTEN && |
| 368 | !between(seq, snd_una, tp->snd_nxt)) { |
| 369 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
| 370 | goto out; |
| 371 | } |
| 372 | |
| 373 | np = inet6_sk(sk); |
| 374 | |
| 375 | if (type == NDISC_REDIRECT) { |
| 376 | struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie); |
| 377 | |
| 378 | if (dst) |
| 379 | dst->ops->redirect(dst, sk, skb); |
| 380 | goto out; |
| 381 | } |
| 382 | |
| 383 | if (type == ICMPV6_PKT_TOOBIG) { |
| 384 | /* We are not interested in TCP_LISTEN and open_requests |
| 385 | * (SYN-ACKs send out by Linux are always <576bytes so |
| 386 | * they should go through unfragmented). |
| 387 | */ |
| 388 | if (sk->sk_state == TCP_LISTEN) |
| 389 | goto out; |
| 390 | |
| 391 | if (!ip6_sk_accept_pmtu(sk)) |
| 392 | goto out; |
| 393 | |
| 394 | tp->mtu_info = ntohl(info); |
| 395 | if (!sock_owned_by_user(sk)) |
| 396 | tcp_v6_mtu_reduced(sk); |
| 397 | else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, |
| 398 | &tp->tsq_flags)) |
| 399 | sock_hold(sk); |
| 400 | goto out; |
| 401 | } |
| 402 | |
| 403 | icmpv6_err_convert(type, code, &err); |
| 404 | |
| 405 | /* Might be for an request_sock */ |
| 406 | switch (sk->sk_state) { |
| 407 | case TCP_SYN_SENT: |
| 408 | case TCP_SYN_RECV: |
| 409 | /* Only in fast or simultaneous open. If a fast open socket is |
| 410 | * is already accepted it is treated as a connected one below. |
| 411 | */ |
| 412 | if (fastopen && !fastopen->sk) |
| 413 | break; |
| 414 | |
| 415 | if (!sock_owned_by_user(sk)) { |
| 416 | sk->sk_err = err; |
| 417 | sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */ |
| 418 | |
| 419 | tcp_done(sk); |
| 420 | } else |
| 421 | sk->sk_err_soft = err; |
| 422 | goto out; |
| 423 | } |
| 424 | |
| 425 | if (!sock_owned_by_user(sk) && np->recverr) { |
| 426 | sk->sk_err = err; |
| 427 | sk->sk_error_report(sk); |
| 428 | } else |
| 429 | sk->sk_err_soft = err; |
| 430 | |
| 431 | out: |
| 432 | bh_unlock_sock(sk); |
| 433 | sock_put(sk); |
| 434 | } |
| 435 | |
| 436 | |
| 437 | static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst, |
| 438 | struct flowi *fl, |
| 439 | struct request_sock *req, |
| 440 | struct tcp_fastopen_cookie *foc, |
| 441 | bool attach_req) |
| 442 | { |
| 443 | struct inet_request_sock *ireq = inet_rsk(req); |
| 444 | struct ipv6_pinfo *np = inet6_sk(sk); |
| 445 | struct flowi6 *fl6 = &fl->u.ip6; |
| 446 | struct sk_buff *skb; |
| 447 | int err = -ENOMEM; |
| 448 | |
| 449 | /* First, grab a route. */ |
| 450 | if (!dst && (dst = inet6_csk_route_req(sk, fl6, req, |
| 451 | IPPROTO_TCP)) == NULL) |
| 452 | goto done; |
| 453 | |
| 454 | skb = tcp_make_synack(sk, dst, req, foc, attach_req); |
| 455 | |
| 456 | if (skb) { |
| 457 | __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr, |
| 458 | &ireq->ir_v6_rmt_addr); |
| 459 | |
| 460 | fl6->daddr = ireq->ir_v6_rmt_addr; |
| 461 | if (np->repflow && ireq->pktopts) |
| 462 | fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts)); |
| 463 | |
| 464 | err = ip6_xmit(sk, skb, fl6, np->opt, np->tclass); |
| 465 | err = net_xmit_eval(err); |
| 466 | } |
| 467 | |
| 468 | done: |
| 469 | return err; |
| 470 | } |
| 471 | |
| 472 | |
| 473 | static void tcp_v6_reqsk_destructor(struct request_sock *req) |
| 474 | { |
| 475 | kfree_skb(inet_rsk(req)->pktopts); |
| 476 | } |
| 477 | |
| 478 | #ifdef CONFIG_TCP_MD5SIG |
| 479 | static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk, |
| 480 | const struct in6_addr *addr) |
| 481 | { |
| 482 | return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6); |
| 483 | } |
| 484 | |
| 485 | static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk, |
| 486 | const struct sock *addr_sk) |
| 487 | { |
| 488 | return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr); |
| 489 | } |
| 490 | |
| 491 | static int tcp_v6_parse_md5_keys(struct sock *sk, char __user *optval, |
| 492 | int optlen) |
| 493 | { |
| 494 | struct tcp_md5sig cmd; |
| 495 | struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr; |
| 496 | |
| 497 | if (optlen < sizeof(cmd)) |
| 498 | return -EINVAL; |
| 499 | |
| 500 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
| 501 | return -EFAULT; |
| 502 | |
| 503 | if (sin6->sin6_family != AF_INET6) |
| 504 | return -EINVAL; |
| 505 | |
| 506 | if (!cmd.tcpm_keylen) { |
| 507 | if (ipv6_addr_v4mapped(&sin6->sin6_addr)) |
| 508 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], |
| 509 | AF_INET); |
| 510 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr, |
| 511 | AF_INET6); |
| 512 | } |
| 513 | |
| 514 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) |
| 515 | return -EINVAL; |
| 516 | |
| 517 | if (ipv6_addr_v4mapped(&sin6->sin6_addr)) |
| 518 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], |
| 519 | AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL); |
| 520 | |
| 521 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr, |
| 522 | AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL); |
| 523 | } |
| 524 | |
| 525 | static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, |
| 526 | const struct in6_addr *daddr, |
| 527 | const struct in6_addr *saddr, int nbytes) |
| 528 | { |
| 529 | struct tcp6_pseudohdr *bp; |
| 530 | struct scatterlist sg; |
| 531 | |
| 532 | bp = &hp->md5_blk.ip6; |
| 533 | /* 1. TCP pseudo-header (RFC2460) */ |
| 534 | bp->saddr = *saddr; |
| 535 | bp->daddr = *daddr; |
| 536 | bp->protocol = cpu_to_be32(IPPROTO_TCP); |
| 537 | bp->len = cpu_to_be32(nbytes); |
| 538 | |
| 539 | sg_init_one(&sg, bp, sizeof(*bp)); |
| 540 | return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); |
| 541 | } |
| 542 | |
| 543 | static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key, |
| 544 | const struct in6_addr *daddr, struct in6_addr *saddr, |
| 545 | const struct tcphdr *th) |
| 546 | { |
| 547 | struct tcp_md5sig_pool *hp; |
| 548 | struct hash_desc *desc; |
| 549 | |
| 550 | hp = tcp_get_md5sig_pool(); |
| 551 | if (!hp) |
| 552 | goto clear_hash_noput; |
| 553 | desc = &hp->md5_desc; |
| 554 | |
| 555 | if (crypto_hash_init(desc)) |
| 556 | goto clear_hash; |
| 557 | if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) |
| 558 | goto clear_hash; |
| 559 | if (tcp_md5_hash_header(hp, th)) |
| 560 | goto clear_hash; |
| 561 | if (tcp_md5_hash_key(hp, key)) |
| 562 | goto clear_hash; |
| 563 | if (crypto_hash_final(desc, md5_hash)) |
| 564 | goto clear_hash; |
| 565 | |
| 566 | tcp_put_md5sig_pool(); |
| 567 | return 0; |
| 568 | |
| 569 | clear_hash: |
| 570 | tcp_put_md5sig_pool(); |
| 571 | clear_hash_noput: |
| 572 | memset(md5_hash, 0, 16); |
| 573 | return 1; |
| 574 | } |
| 575 | |
| 576 | static int tcp_v6_md5_hash_skb(char *md5_hash, |
| 577 | const struct tcp_md5sig_key *key, |
| 578 | const struct sock *sk, |
| 579 | const struct sk_buff *skb) |
| 580 | { |
| 581 | const struct in6_addr *saddr, *daddr; |
| 582 | struct tcp_md5sig_pool *hp; |
| 583 | struct hash_desc *desc; |
| 584 | const struct tcphdr *th = tcp_hdr(skb); |
| 585 | |
| 586 | if (sk) { /* valid for establish/request sockets */ |
| 587 | saddr = &sk->sk_v6_rcv_saddr; |
| 588 | daddr = &sk->sk_v6_daddr; |
| 589 | } else { |
| 590 | const struct ipv6hdr *ip6h = ipv6_hdr(skb); |
| 591 | saddr = &ip6h->saddr; |
| 592 | daddr = &ip6h->daddr; |
| 593 | } |
| 594 | |
| 595 | hp = tcp_get_md5sig_pool(); |
| 596 | if (!hp) |
| 597 | goto clear_hash_noput; |
| 598 | desc = &hp->md5_desc; |
| 599 | |
| 600 | if (crypto_hash_init(desc)) |
| 601 | goto clear_hash; |
| 602 | |
| 603 | if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) |
| 604 | goto clear_hash; |
| 605 | if (tcp_md5_hash_header(hp, th)) |
| 606 | goto clear_hash; |
| 607 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) |
| 608 | goto clear_hash; |
| 609 | if (tcp_md5_hash_key(hp, key)) |
| 610 | goto clear_hash; |
| 611 | if (crypto_hash_final(desc, md5_hash)) |
| 612 | goto clear_hash; |
| 613 | |
| 614 | tcp_put_md5sig_pool(); |
| 615 | return 0; |
| 616 | |
| 617 | clear_hash: |
| 618 | tcp_put_md5sig_pool(); |
| 619 | clear_hash_noput: |
| 620 | memset(md5_hash, 0, 16); |
| 621 | return 1; |
| 622 | } |
| 623 | |
| 624 | #endif |
| 625 | |
| 626 | static bool tcp_v6_inbound_md5_hash(const struct sock *sk, |
| 627 | const struct sk_buff *skb) |
| 628 | { |
| 629 | #ifdef CONFIG_TCP_MD5SIG |
| 630 | const __u8 *hash_location = NULL; |
| 631 | struct tcp_md5sig_key *hash_expected; |
| 632 | const struct ipv6hdr *ip6h = ipv6_hdr(skb); |
| 633 | const struct tcphdr *th = tcp_hdr(skb); |
| 634 | int genhash; |
| 635 | u8 newhash[16]; |
| 636 | |
| 637 | hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr); |
| 638 | hash_location = tcp_parse_md5sig_option(th); |
| 639 | |
| 640 | /* We've parsed the options - do we have a hash? */ |
| 641 | if (!hash_expected && !hash_location) |
| 642 | return false; |
| 643 | |
| 644 | if (hash_expected && !hash_location) { |
| 645 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
| 646 | return true; |
| 647 | } |
| 648 | |
| 649 | if (!hash_expected && hash_location) { |
| 650 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
| 651 | return true; |
| 652 | } |
| 653 | |
| 654 | /* check the signature */ |
| 655 | genhash = tcp_v6_md5_hash_skb(newhash, |
| 656 | hash_expected, |
| 657 | NULL, skb); |
| 658 | |
| 659 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { |
| 660 | net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n", |
| 661 | genhash ? "failed" : "mismatch", |
| 662 | &ip6h->saddr, ntohs(th->source), |
| 663 | &ip6h->daddr, ntohs(th->dest)); |
| 664 | return true; |
| 665 | } |
| 666 | #endif |
| 667 | return false; |
| 668 | } |
| 669 | |
| 670 | static void tcp_v6_init_req(struct request_sock *req, |
| 671 | const struct sock *sk_listener, |
| 672 | struct sk_buff *skb) |
| 673 | { |
| 674 | struct inet_request_sock *ireq = inet_rsk(req); |
| 675 | const struct ipv6_pinfo *np = inet6_sk(sk_listener); |
| 676 | |
| 677 | ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; |
| 678 | ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; |
| 679 | |
| 680 | /* So that link locals have meaning */ |
| 681 | if (!sk_listener->sk_bound_dev_if && |
| 682 | ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) |
| 683 | ireq->ir_iif = tcp_v6_iif(skb); |
| 684 | |
| 685 | if (!TCP_SKB_CB(skb)->tcp_tw_isn && |
| 686 | (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) || |
| 687 | np->rxopt.bits.rxinfo || |
| 688 | np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim || |
| 689 | np->rxopt.bits.rxohlim || np->repflow)) { |
| 690 | atomic_inc(&skb->users); |
| 691 | ireq->pktopts = skb; |
| 692 | } |
| 693 | } |
| 694 | |
| 695 | static struct dst_entry *tcp_v6_route_req(const struct sock *sk, |
| 696 | struct flowi *fl, |
| 697 | const struct request_sock *req, |
| 698 | bool *strict) |
| 699 | { |
| 700 | if (strict) |
| 701 | *strict = true; |
| 702 | return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP); |
| 703 | } |
| 704 | |
| 705 | struct request_sock_ops tcp6_request_sock_ops __read_mostly = { |
| 706 | .family = AF_INET6, |
| 707 | .obj_size = sizeof(struct tcp6_request_sock), |
| 708 | .rtx_syn_ack = tcp_rtx_synack, |
| 709 | .send_ack = tcp_v6_reqsk_send_ack, |
| 710 | .destructor = tcp_v6_reqsk_destructor, |
| 711 | .send_reset = tcp_v6_send_reset, |
| 712 | .syn_ack_timeout = tcp_syn_ack_timeout, |
| 713 | }; |
| 714 | |
| 715 | static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { |
| 716 | .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - |
| 717 | sizeof(struct ipv6hdr), |
| 718 | #ifdef CONFIG_TCP_MD5SIG |
| 719 | .req_md5_lookup = tcp_v6_md5_lookup, |
| 720 | .calc_md5_hash = tcp_v6_md5_hash_skb, |
| 721 | #endif |
| 722 | .init_req = tcp_v6_init_req, |
| 723 | #ifdef CONFIG_SYN_COOKIES |
| 724 | .cookie_init_seq = cookie_v6_init_sequence, |
| 725 | #endif |
| 726 | .route_req = tcp_v6_route_req, |
| 727 | .init_seq = tcp_v6_init_sequence, |
| 728 | .send_synack = tcp_v6_send_synack, |
| 729 | }; |
| 730 | |
| 731 | static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq, |
| 732 | u32 ack, u32 win, u32 tsval, u32 tsecr, |
| 733 | int oif, struct tcp_md5sig_key *key, int rst, |
| 734 | u8 tclass, u32 label) |
| 735 | { |
| 736 | const struct tcphdr *th = tcp_hdr(skb); |
| 737 | struct tcphdr *t1; |
| 738 | struct sk_buff *buff; |
| 739 | struct flowi6 fl6; |
| 740 | struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); |
| 741 | struct sock *ctl_sk = net->ipv6.tcp_sk; |
| 742 | unsigned int tot_len = sizeof(struct tcphdr); |
| 743 | struct dst_entry *dst; |
| 744 | __be32 *topt; |
| 745 | |
| 746 | if (tsecr) |
| 747 | tot_len += TCPOLEN_TSTAMP_ALIGNED; |
| 748 | #ifdef CONFIG_TCP_MD5SIG |
| 749 | if (key) |
| 750 | tot_len += TCPOLEN_MD5SIG_ALIGNED; |
| 751 | #endif |
| 752 | |
| 753 | buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len, |
| 754 | GFP_ATOMIC); |
| 755 | if (!buff) |
| 756 | return; |
| 757 | |
| 758 | skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len); |
| 759 | |
| 760 | t1 = (struct tcphdr *) skb_push(buff, tot_len); |
| 761 | skb_reset_transport_header(buff); |
| 762 | |
| 763 | /* Swap the send and the receive. */ |
| 764 | memset(t1, 0, sizeof(*t1)); |
| 765 | t1->dest = th->source; |
| 766 | t1->source = th->dest; |
| 767 | t1->doff = tot_len / 4; |
| 768 | t1->seq = htonl(seq); |
| 769 | t1->ack_seq = htonl(ack); |
| 770 | t1->ack = !rst || !th->ack; |
| 771 | t1->rst = rst; |
| 772 | t1->window = htons(win); |
| 773 | |
| 774 | topt = (__be32 *)(t1 + 1); |
| 775 | |
| 776 | if (tsecr) { |
| 777 | *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
| 778 | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); |
| 779 | *topt++ = htonl(tsval); |
| 780 | *topt++ = htonl(tsecr); |
| 781 | } |
| 782 | |
| 783 | #ifdef CONFIG_TCP_MD5SIG |
| 784 | if (key) { |
| 785 | *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
| 786 | (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); |
| 787 | tcp_v6_md5_hash_hdr((__u8 *)topt, key, |
| 788 | &ipv6_hdr(skb)->saddr, |
| 789 | &ipv6_hdr(skb)->daddr, t1); |
| 790 | } |
| 791 | #endif |
| 792 | |
| 793 | memset(&fl6, 0, sizeof(fl6)); |
| 794 | fl6.daddr = ipv6_hdr(skb)->saddr; |
| 795 | fl6.saddr = ipv6_hdr(skb)->daddr; |
| 796 | fl6.flowlabel = label; |
| 797 | |
| 798 | buff->ip_summed = CHECKSUM_PARTIAL; |
| 799 | buff->csum = 0; |
| 800 | |
| 801 | __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr); |
| 802 | |
| 803 | fl6.flowi6_proto = IPPROTO_TCP; |
| 804 | if (rt6_need_strict(&fl6.daddr) && !oif) |
| 805 | fl6.flowi6_oif = tcp_v6_iif(skb); |
| 806 | else |
| 807 | fl6.flowi6_oif = oif; |
| 808 | fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark); |
| 809 | fl6.fl6_dport = t1->dest; |
| 810 | fl6.fl6_sport = t1->source; |
| 811 | security_skb_classify_flow(skb, flowi6_to_flowi(&fl6)); |
| 812 | |
| 813 | /* Pass a socket to ip6_dst_lookup either it is for RST |
| 814 | * Underlying function will use this to retrieve the network |
| 815 | * namespace |
| 816 | */ |
| 817 | dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL); |
| 818 | if (!IS_ERR(dst)) { |
| 819 | skb_dst_set(buff, dst); |
| 820 | ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass); |
| 821 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
| 822 | if (rst) |
| 823 | TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); |
| 824 | return; |
| 825 | } |
| 826 | |
| 827 | kfree_skb(buff); |
| 828 | } |
| 829 | |
| 830 | static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb) |
| 831 | { |
| 832 | const struct tcphdr *th = tcp_hdr(skb); |
| 833 | u32 seq = 0, ack_seq = 0; |
| 834 | struct tcp_md5sig_key *key = NULL; |
| 835 | #ifdef CONFIG_TCP_MD5SIG |
| 836 | const __u8 *hash_location = NULL; |
| 837 | struct ipv6hdr *ipv6h = ipv6_hdr(skb); |
| 838 | unsigned char newhash[16]; |
| 839 | int genhash; |
| 840 | struct sock *sk1 = NULL; |
| 841 | #endif |
| 842 | int oif; |
| 843 | |
| 844 | if (th->rst) |
| 845 | return; |
| 846 | |
| 847 | /* If sk not NULL, it means we did a successful lookup and incoming |
| 848 | * route had to be correct. prequeue might have dropped our dst. |
| 849 | */ |
| 850 | if (!sk && !ipv6_unicast_destination(skb)) |
| 851 | return; |
| 852 | |
| 853 | #ifdef CONFIG_TCP_MD5SIG |
| 854 | hash_location = tcp_parse_md5sig_option(th); |
| 855 | if (!sk && hash_location) { |
| 856 | /* |
| 857 | * active side is lost. Try to find listening socket through |
| 858 | * source port, and then find md5 key through listening socket. |
| 859 | * we are not loose security here: |
| 860 | * Incoming packet is checked with md5 hash with finding key, |
| 861 | * no RST generated if md5 hash doesn't match. |
| 862 | */ |
| 863 | sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev), |
| 864 | &tcp_hashinfo, &ipv6h->saddr, |
| 865 | th->source, &ipv6h->daddr, |
| 866 | ntohs(th->source), tcp_v6_iif(skb)); |
| 867 | if (!sk1) |
| 868 | return; |
| 869 | |
| 870 | rcu_read_lock(); |
| 871 | key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr); |
| 872 | if (!key) |
| 873 | goto release_sk1; |
| 874 | |
| 875 | genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb); |
| 876 | if (genhash || memcmp(hash_location, newhash, 16) != 0) |
| 877 | goto release_sk1; |
| 878 | } else { |
| 879 | key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL; |
| 880 | } |
| 881 | #endif |
| 882 | |
| 883 | if (th->ack) |
| 884 | seq = ntohl(th->ack_seq); |
| 885 | else |
| 886 | ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - |
| 887 | (th->doff << 2); |
| 888 | |
| 889 | oif = sk ? sk->sk_bound_dev_if : 0; |
| 890 | tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1, 0, 0); |
| 891 | |
| 892 | #ifdef CONFIG_TCP_MD5SIG |
| 893 | release_sk1: |
| 894 | if (sk1) { |
| 895 | rcu_read_unlock(); |
| 896 | sock_put(sk1); |
| 897 | } |
| 898 | #endif |
| 899 | } |
| 900 | |
| 901 | static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, |
| 902 | u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, |
| 903 | struct tcp_md5sig_key *key, u8 tclass, |
| 904 | u32 label) |
| 905 | { |
| 906 | tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0, |
| 907 | tclass, label); |
| 908 | } |
| 909 | |
| 910 | static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb) |
| 911 | { |
| 912 | struct inet_timewait_sock *tw = inet_twsk(sk); |
| 913 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
| 914 | |
| 915 | tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
| 916 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
| 917 | tcp_time_stamp + tcptw->tw_ts_offset, |
| 918 | tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw), |
| 919 | tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel)); |
| 920 | |
| 921 | inet_twsk_put(tw); |
| 922 | } |
| 923 | |
| 924 | static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, |
| 925 | struct request_sock *req) |
| 926 | { |
| 927 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV |
| 928 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. |
| 929 | */ |
| 930 | tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ? |
| 931 | tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, |
| 932 | tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd, |
| 933 | tcp_time_stamp, req->ts_recent, sk->sk_bound_dev_if, |
| 934 | tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), |
| 935 | 0, 0); |
| 936 | } |
| 937 | |
| 938 | |
| 939 | static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb) |
| 940 | { |
| 941 | #ifdef CONFIG_SYN_COOKIES |
| 942 | const struct tcphdr *th = tcp_hdr(skb); |
| 943 | |
| 944 | if (!th->syn) |
| 945 | sk = cookie_v6_check(sk, skb); |
| 946 | #endif |
| 947 | return sk; |
| 948 | } |
| 949 | |
| 950 | static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) |
| 951 | { |
| 952 | if (skb->protocol == htons(ETH_P_IP)) |
| 953 | return tcp_v4_conn_request(sk, skb); |
| 954 | |
| 955 | if (!ipv6_unicast_destination(skb)) |
| 956 | goto drop; |
| 957 | |
| 958 | return tcp_conn_request(&tcp6_request_sock_ops, |
| 959 | &tcp_request_sock_ipv6_ops, sk, skb); |
| 960 | |
| 961 | drop: |
| 962 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
| 963 | return 0; /* don't send reset */ |
| 964 | } |
| 965 | |
| 966 | static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, |
| 967 | struct request_sock *req, |
| 968 | struct dst_entry *dst, |
| 969 | struct request_sock *req_unhash, |
| 970 | bool *own_req) |
| 971 | { |
| 972 | struct inet_request_sock *ireq; |
| 973 | struct ipv6_pinfo *newnp; |
| 974 | const struct ipv6_pinfo *np = inet6_sk(sk); |
| 975 | struct tcp6_sock *newtcp6sk; |
| 976 | struct inet_sock *newinet; |
| 977 | struct tcp_sock *newtp; |
| 978 | struct sock *newsk; |
| 979 | #ifdef CONFIG_TCP_MD5SIG |
| 980 | struct tcp_md5sig_key *key; |
| 981 | #endif |
| 982 | struct flowi6 fl6; |
| 983 | |
| 984 | if (skb->protocol == htons(ETH_P_IP)) { |
| 985 | /* |
| 986 | * v6 mapped |
| 987 | */ |
| 988 | |
| 989 | newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst, |
| 990 | req_unhash, own_req); |
| 991 | |
| 992 | if (!newsk) |
| 993 | return NULL; |
| 994 | |
| 995 | newtcp6sk = (struct tcp6_sock *)newsk; |
| 996 | inet_sk(newsk)->pinet6 = &newtcp6sk->inet6; |
| 997 | |
| 998 | newinet = inet_sk(newsk); |
| 999 | newnp = inet6_sk(newsk); |
| 1000 | newtp = tcp_sk(newsk); |
| 1001 | |
| 1002 | memcpy(newnp, np, sizeof(struct ipv6_pinfo)); |
| 1003 | |
| 1004 | newnp->saddr = newsk->sk_v6_rcv_saddr; |
| 1005 | |
| 1006 | inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; |
| 1007 | newsk->sk_backlog_rcv = tcp_v4_do_rcv; |
| 1008 | #ifdef CONFIG_TCP_MD5SIG |
| 1009 | newtp->af_specific = &tcp_sock_ipv6_mapped_specific; |
| 1010 | #endif |
| 1011 | |
| 1012 | newnp->ipv6_ac_list = NULL; |
| 1013 | newnp->ipv6_fl_list = NULL; |
| 1014 | newnp->pktoptions = NULL; |
| 1015 | newnp->opt = NULL; |
| 1016 | newnp->mcast_oif = tcp_v6_iif(skb); |
| 1017 | newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; |
| 1018 | newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb)); |
| 1019 | if (np->repflow) |
| 1020 | newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb)); |
| 1021 | |
| 1022 | /* |
| 1023 | * No need to charge this sock to the relevant IPv6 refcnt debug socks count |
| 1024 | * here, tcp_create_openreq_child now does this for us, see the comment in |
| 1025 | * that function for the gory details. -acme |
| 1026 | */ |
| 1027 | |
| 1028 | /* It is tricky place. Until this moment IPv4 tcp |
| 1029 | worked with IPv6 icsk.icsk_af_ops. |
| 1030 | Sync it now. |
| 1031 | */ |
| 1032 | tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie); |
| 1033 | |
| 1034 | return newsk; |
| 1035 | } |
| 1036 | |
| 1037 | ireq = inet_rsk(req); |
| 1038 | |
| 1039 | if (sk_acceptq_is_full(sk)) |
| 1040 | goto out_overflow; |
| 1041 | |
| 1042 | if (!dst) { |
| 1043 | dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP); |
| 1044 | if (!dst) |
| 1045 | goto out; |
| 1046 | } |
| 1047 | |
| 1048 | newsk = tcp_create_openreq_child(sk, req, skb); |
| 1049 | if (!newsk) |
| 1050 | goto out_nonewsk; |
| 1051 | |
| 1052 | /* |
| 1053 | * No need to charge this sock to the relevant IPv6 refcnt debug socks |
| 1054 | * count here, tcp_create_openreq_child now does this for us, see the |
| 1055 | * comment in that function for the gory details. -acme |
| 1056 | */ |
| 1057 | |
| 1058 | newsk->sk_gso_type = SKB_GSO_TCPV6; |
| 1059 | __ip6_dst_store(newsk, dst, NULL, NULL); |
| 1060 | inet6_sk_rx_dst_set(newsk, skb); |
| 1061 | |
| 1062 | newtcp6sk = (struct tcp6_sock *)newsk; |
| 1063 | inet_sk(newsk)->pinet6 = &newtcp6sk->inet6; |
| 1064 | |
| 1065 | newtp = tcp_sk(newsk); |
| 1066 | newinet = inet_sk(newsk); |
| 1067 | newnp = inet6_sk(newsk); |
| 1068 | |
| 1069 | memcpy(newnp, np, sizeof(struct ipv6_pinfo)); |
| 1070 | |
| 1071 | newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr; |
| 1072 | newnp->saddr = ireq->ir_v6_loc_addr; |
| 1073 | newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr; |
| 1074 | newsk->sk_bound_dev_if = ireq->ir_iif; |
| 1075 | |
| 1076 | /* Now IPv6 options... |
| 1077 | |
| 1078 | First: no IPv4 options. |
| 1079 | */ |
| 1080 | newinet->inet_opt = NULL; |
| 1081 | newnp->ipv6_ac_list = NULL; |
| 1082 | newnp->ipv6_fl_list = NULL; |
| 1083 | |
| 1084 | /* Clone RX bits */ |
| 1085 | newnp->rxopt.all = np->rxopt.all; |
| 1086 | |
| 1087 | newnp->pktoptions = NULL; |
| 1088 | newnp->opt = NULL; |
| 1089 | newnp->mcast_oif = tcp_v6_iif(skb); |
| 1090 | newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; |
| 1091 | newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb)); |
| 1092 | if (np->repflow) |
| 1093 | newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb)); |
| 1094 | |
| 1095 | /* Clone native IPv6 options from listening socket (if any) |
| 1096 | |
| 1097 | Yes, keeping reference count would be much more clever, |
| 1098 | but we make one more one thing there: reattach optmem |
| 1099 | to newsk. |
| 1100 | */ |
| 1101 | if (np->opt) |
| 1102 | newnp->opt = ipv6_dup_options(newsk, np->opt); |
| 1103 | |
| 1104 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
| 1105 | if (newnp->opt) |
| 1106 | inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen + |
| 1107 | newnp->opt->opt_flen); |
| 1108 | |
| 1109 | tcp_ca_openreq_child(newsk, dst); |
| 1110 | |
| 1111 | tcp_sync_mss(newsk, dst_mtu(dst)); |
| 1112 | newtp->advmss = dst_metric_advmss(dst); |
| 1113 | if (tcp_sk(sk)->rx_opt.user_mss && |
| 1114 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) |
| 1115 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; |
| 1116 | |
| 1117 | tcp_initialize_rcv_mss(newsk); |
| 1118 | |
| 1119 | newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6; |
| 1120 | newinet->inet_rcv_saddr = LOOPBACK4_IPV6; |
| 1121 | |
| 1122 | #ifdef CONFIG_TCP_MD5SIG |
| 1123 | /* Copy over the MD5 key from the original socket */ |
| 1124 | key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr); |
| 1125 | if (key) { |
| 1126 | /* We're using one, so create a matching key |
| 1127 | * on the newsk structure. If we fail to get |
| 1128 | * memory, then we end up not copying the key |
| 1129 | * across. Shucks. |
| 1130 | */ |
| 1131 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr, |
| 1132 | AF_INET6, key->key, key->keylen, |
| 1133 | sk_gfp_atomic(sk, GFP_ATOMIC)); |
| 1134 | } |
| 1135 | #endif |
| 1136 | |
| 1137 | if (__inet_inherit_port(sk, newsk) < 0) { |
| 1138 | inet_csk_prepare_forced_close(newsk); |
| 1139 | tcp_done(newsk); |
| 1140 | goto out; |
| 1141 | } |
| 1142 | *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash)); |
| 1143 | /* Clone pktoptions received with SYN, if we own the req */ |
| 1144 | if (*own_req && ireq->pktopts) { |
| 1145 | newnp->pktoptions = skb_clone(ireq->pktopts, |
| 1146 | sk_gfp_atomic(sk, GFP_ATOMIC)); |
| 1147 | consume_skb(ireq->pktopts); |
| 1148 | ireq->pktopts = NULL; |
| 1149 | if (newnp->pktoptions) |
| 1150 | skb_set_owner_r(newnp->pktoptions, newsk); |
| 1151 | } |
| 1152 | |
| 1153 | return newsk; |
| 1154 | |
| 1155 | out_overflow: |
| 1156 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
| 1157 | out_nonewsk: |
| 1158 | dst_release(dst); |
| 1159 | out: |
| 1160 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
| 1161 | return NULL; |
| 1162 | } |
| 1163 | |
| 1164 | /* The socket must have it's spinlock held when we get |
| 1165 | * here, unless it is a TCP_LISTEN socket. |
| 1166 | * |
| 1167 | * We have a potential double-lock case here, so even when |
| 1168 | * doing backlog processing we use the BH locking scheme. |
| 1169 | * This is because we cannot sleep with the original spinlock |
| 1170 | * held. |
| 1171 | */ |
| 1172 | static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) |
| 1173 | { |
| 1174 | struct ipv6_pinfo *np = inet6_sk(sk); |
| 1175 | struct tcp_sock *tp; |
| 1176 | struct sk_buff *opt_skb = NULL; |
| 1177 | |
| 1178 | /* Imagine: socket is IPv6. IPv4 packet arrives, |
| 1179 | goes to IPv4 receive handler and backlogged. |
| 1180 | From backlog it always goes here. Kerboom... |
| 1181 | Fortunately, tcp_rcv_established and rcv_established |
| 1182 | handle them correctly, but it is not case with |
| 1183 | tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK |
| 1184 | */ |
| 1185 | |
| 1186 | if (skb->protocol == htons(ETH_P_IP)) |
| 1187 | return tcp_v4_do_rcv(sk, skb); |
| 1188 | |
| 1189 | if (sk_filter(sk, skb)) |
| 1190 | goto discard; |
| 1191 | |
| 1192 | /* |
| 1193 | * socket locking is here for SMP purposes as backlog rcv |
| 1194 | * is currently called with bh processing disabled. |
| 1195 | */ |
| 1196 | |
| 1197 | /* Do Stevens' IPV6_PKTOPTIONS. |
| 1198 | |
| 1199 | Yes, guys, it is the only place in our code, where we |
| 1200 | may make it not affecting IPv4. |
| 1201 | The rest of code is protocol independent, |
| 1202 | and I do not like idea to uglify IPv4. |
| 1203 | |
| 1204 | Actually, all the idea behind IPV6_PKTOPTIONS |
| 1205 | looks not very well thought. For now we latch |
| 1206 | options, received in the last packet, enqueued |
| 1207 | by tcp. Feel free to propose better solution. |
| 1208 | --ANK (980728) |
| 1209 | */ |
| 1210 | if (np->rxopt.all) |
| 1211 | opt_skb = skb_clone(skb, sk_gfp_atomic(sk, GFP_ATOMIC)); |
| 1212 | |
| 1213 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
| 1214 | struct dst_entry *dst = sk->sk_rx_dst; |
| 1215 | |
| 1216 | sock_rps_save_rxhash(sk, skb); |
| 1217 | sk_mark_napi_id(sk, skb); |
| 1218 | if (dst) { |
| 1219 | if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || |
| 1220 | dst->ops->check(dst, np->rx_dst_cookie) == NULL) { |
| 1221 | dst_release(dst); |
| 1222 | sk->sk_rx_dst = NULL; |
| 1223 | } |
| 1224 | } |
| 1225 | |
| 1226 | tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len); |
| 1227 | if (opt_skb) |
| 1228 | goto ipv6_pktoptions; |
| 1229 | return 0; |
| 1230 | } |
| 1231 | |
| 1232 | if (tcp_checksum_complete(skb)) |
| 1233 | goto csum_err; |
| 1234 | |
| 1235 | if (sk->sk_state == TCP_LISTEN) { |
| 1236 | struct sock *nsk = tcp_v6_cookie_check(sk, skb); |
| 1237 | |
| 1238 | if (!nsk) |
| 1239 | goto discard; |
| 1240 | |
| 1241 | if (nsk != sk) { |
| 1242 | sock_rps_save_rxhash(nsk, skb); |
| 1243 | sk_mark_napi_id(nsk, skb); |
| 1244 | if (tcp_child_process(sk, nsk, skb)) |
| 1245 | goto reset; |
| 1246 | if (opt_skb) |
| 1247 | __kfree_skb(opt_skb); |
| 1248 | return 0; |
| 1249 | } |
| 1250 | } else |
| 1251 | sock_rps_save_rxhash(sk, skb); |
| 1252 | |
| 1253 | if (tcp_rcv_state_process(sk, skb)) |
| 1254 | goto reset; |
| 1255 | if (opt_skb) |
| 1256 | goto ipv6_pktoptions; |
| 1257 | return 0; |
| 1258 | |
| 1259 | reset: |
| 1260 | tcp_v6_send_reset(sk, skb); |
| 1261 | discard: |
| 1262 | if (opt_skb) |
| 1263 | __kfree_skb(opt_skb); |
| 1264 | kfree_skb(skb); |
| 1265 | return 0; |
| 1266 | csum_err: |
| 1267 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); |
| 1268 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
| 1269 | goto discard; |
| 1270 | |
| 1271 | |
| 1272 | ipv6_pktoptions: |
| 1273 | /* Do you ask, what is it? |
| 1274 | |
| 1275 | 1. skb was enqueued by tcp. |
| 1276 | 2. skb is added to tail of read queue, rather than out of order. |
| 1277 | 3. socket is not in passive state. |
| 1278 | 4. Finally, it really contains options, which user wants to receive. |
| 1279 | */ |
| 1280 | tp = tcp_sk(sk); |
| 1281 | if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && |
| 1282 | !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { |
| 1283 | if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) |
| 1284 | np->mcast_oif = tcp_v6_iif(opt_skb); |
| 1285 | if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) |
| 1286 | np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit; |
| 1287 | if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass) |
| 1288 | np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb)); |
| 1289 | if (np->repflow) |
| 1290 | np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb)); |
| 1291 | if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) { |
| 1292 | skb_set_owner_r(opt_skb, sk); |
| 1293 | opt_skb = xchg(&np->pktoptions, opt_skb); |
| 1294 | } else { |
| 1295 | __kfree_skb(opt_skb); |
| 1296 | opt_skb = xchg(&np->pktoptions, NULL); |
| 1297 | } |
| 1298 | } |
| 1299 | |
| 1300 | kfree_skb(opt_skb); |
| 1301 | return 0; |
| 1302 | } |
| 1303 | |
| 1304 | static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr, |
| 1305 | const struct tcphdr *th) |
| 1306 | { |
| 1307 | /* This is tricky: we move IP6CB at its correct location into |
| 1308 | * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because |
| 1309 | * _decode_session6() uses IP6CB(). |
| 1310 | * barrier() makes sure compiler won't play aliasing games. |
| 1311 | */ |
| 1312 | memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb), |
| 1313 | sizeof(struct inet6_skb_parm)); |
| 1314 | barrier(); |
| 1315 | |
| 1316 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
| 1317 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + |
| 1318 | skb->len - th->doff*4); |
| 1319 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); |
| 1320 | TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); |
| 1321 | TCP_SKB_CB(skb)->tcp_tw_isn = 0; |
| 1322 | TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr); |
| 1323 | TCP_SKB_CB(skb)->sacked = 0; |
| 1324 | } |
| 1325 | |
| 1326 | static void tcp_v6_restore_cb(struct sk_buff *skb) |
| 1327 | { |
| 1328 | /* We need to move header back to the beginning if xfrm6_policy_check() |
| 1329 | * and tcp_v6_fill_cb() are going to be called again. |
| 1330 | */ |
| 1331 | memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6, |
| 1332 | sizeof(struct inet6_skb_parm)); |
| 1333 | } |
| 1334 | |
| 1335 | static int tcp_v6_rcv(struct sk_buff *skb) |
| 1336 | { |
| 1337 | const struct tcphdr *th; |
| 1338 | const struct ipv6hdr *hdr; |
| 1339 | struct sock *sk; |
| 1340 | int ret; |
| 1341 | struct net *net = dev_net(skb->dev); |
| 1342 | |
| 1343 | if (skb->pkt_type != PACKET_HOST) |
| 1344 | goto discard_it; |
| 1345 | |
| 1346 | /* |
| 1347 | * Count it even if it's bad. |
| 1348 | */ |
| 1349 | TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); |
| 1350 | |
| 1351 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) |
| 1352 | goto discard_it; |
| 1353 | |
| 1354 | th = tcp_hdr(skb); |
| 1355 | |
| 1356 | if (th->doff < sizeof(struct tcphdr)/4) |
| 1357 | goto bad_packet; |
| 1358 | if (!pskb_may_pull(skb, th->doff*4)) |
| 1359 | goto discard_it; |
| 1360 | |
| 1361 | if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo)) |
| 1362 | goto csum_error; |
| 1363 | |
| 1364 | th = tcp_hdr(skb); |
| 1365 | hdr = ipv6_hdr(skb); |
| 1366 | |
| 1367 | lookup: |
| 1368 | sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest, |
| 1369 | inet6_iif(skb)); |
| 1370 | if (!sk) |
| 1371 | goto no_tcp_socket; |
| 1372 | |
| 1373 | process: |
| 1374 | if (sk->sk_state == TCP_TIME_WAIT) |
| 1375 | goto do_time_wait; |
| 1376 | |
| 1377 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
| 1378 | struct request_sock *req = inet_reqsk(sk); |
| 1379 | struct sock *nsk = NULL; |
| 1380 | |
| 1381 | sk = req->rsk_listener; |
| 1382 | tcp_v6_fill_cb(skb, hdr, th); |
| 1383 | if (tcp_v6_inbound_md5_hash(sk, skb)) { |
| 1384 | reqsk_put(req); |
| 1385 | goto discard_it; |
| 1386 | } |
| 1387 | if (likely(sk->sk_state == TCP_LISTEN)) { |
| 1388 | nsk = tcp_check_req(sk, skb, req, false); |
| 1389 | } else { |
| 1390 | inet_csk_reqsk_queue_drop_and_put(sk, req); |
| 1391 | goto lookup; |
| 1392 | } |
| 1393 | if (!nsk) { |
| 1394 | reqsk_put(req); |
| 1395 | goto discard_it; |
| 1396 | } |
| 1397 | if (nsk == sk) { |
| 1398 | sock_hold(sk); |
| 1399 | reqsk_put(req); |
| 1400 | tcp_v6_restore_cb(skb); |
| 1401 | } else if (tcp_child_process(sk, nsk, skb)) { |
| 1402 | tcp_v6_send_reset(nsk, skb); |
| 1403 | goto discard_it; |
| 1404 | } else { |
| 1405 | return 0; |
| 1406 | } |
| 1407 | } |
| 1408 | if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) { |
| 1409 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); |
| 1410 | goto discard_and_relse; |
| 1411 | } |
| 1412 | |
| 1413 | if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) |
| 1414 | goto discard_and_relse; |
| 1415 | |
| 1416 | tcp_v6_fill_cb(skb, hdr, th); |
| 1417 | |
| 1418 | if (tcp_v6_inbound_md5_hash(sk, skb)) |
| 1419 | goto discard_and_relse; |
| 1420 | |
| 1421 | if (sk_filter(sk, skb)) |
| 1422 | goto discard_and_relse; |
| 1423 | |
| 1424 | skb->dev = NULL; |
| 1425 | |
| 1426 | if (sk->sk_state == TCP_LISTEN) { |
| 1427 | ret = tcp_v6_do_rcv(sk, skb); |
| 1428 | goto put_and_return; |
| 1429 | } |
| 1430 | |
| 1431 | sk_incoming_cpu_update(sk); |
| 1432 | |
| 1433 | bh_lock_sock_nested(sk); |
| 1434 | tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs); |
| 1435 | ret = 0; |
| 1436 | if (!sock_owned_by_user(sk)) { |
| 1437 | if (!tcp_prequeue(sk, skb)) |
| 1438 | ret = tcp_v6_do_rcv(sk, skb); |
| 1439 | } else if (unlikely(sk_add_backlog(sk, skb, |
| 1440 | sk->sk_rcvbuf + sk->sk_sndbuf))) { |
| 1441 | bh_unlock_sock(sk); |
| 1442 | NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); |
| 1443 | goto discard_and_relse; |
| 1444 | } |
| 1445 | bh_unlock_sock(sk); |
| 1446 | |
| 1447 | put_and_return: |
| 1448 | sock_put(sk); |
| 1449 | return ret ? -1 : 0; |
| 1450 | |
| 1451 | no_tcp_socket: |
| 1452 | if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) |
| 1453 | goto discard_it; |
| 1454 | |
| 1455 | tcp_v6_fill_cb(skb, hdr, th); |
| 1456 | |
| 1457 | if (tcp_checksum_complete(skb)) { |
| 1458 | csum_error: |
| 1459 | TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS); |
| 1460 | bad_packet: |
| 1461 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
| 1462 | } else { |
| 1463 | tcp_v6_send_reset(NULL, skb); |
| 1464 | } |
| 1465 | |
| 1466 | discard_it: |
| 1467 | kfree_skb(skb); |
| 1468 | return 0; |
| 1469 | |
| 1470 | discard_and_relse: |
| 1471 | sock_put(sk); |
| 1472 | goto discard_it; |
| 1473 | |
| 1474 | do_time_wait: |
| 1475 | if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) { |
| 1476 | inet_twsk_put(inet_twsk(sk)); |
| 1477 | goto discard_it; |
| 1478 | } |
| 1479 | |
| 1480 | tcp_v6_fill_cb(skb, hdr, th); |
| 1481 | |
| 1482 | if (tcp_checksum_complete(skb)) { |
| 1483 | inet_twsk_put(inet_twsk(sk)); |
| 1484 | goto csum_error; |
| 1485 | } |
| 1486 | |
| 1487 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
| 1488 | case TCP_TW_SYN: |
| 1489 | { |
| 1490 | struct sock *sk2; |
| 1491 | |
| 1492 | sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo, |
| 1493 | &ipv6_hdr(skb)->saddr, th->source, |
| 1494 | &ipv6_hdr(skb)->daddr, |
| 1495 | ntohs(th->dest), tcp_v6_iif(skb)); |
| 1496 | if (sk2) { |
| 1497 | struct inet_timewait_sock *tw = inet_twsk(sk); |
| 1498 | inet_twsk_deschedule_put(tw); |
| 1499 | sk = sk2; |
| 1500 | tcp_v6_restore_cb(skb); |
| 1501 | goto process; |
| 1502 | } |
| 1503 | /* Fall through to ACK */ |
| 1504 | } |
| 1505 | case TCP_TW_ACK: |
| 1506 | tcp_v6_timewait_ack(sk, skb); |
| 1507 | break; |
| 1508 | case TCP_TW_RST: |
| 1509 | tcp_v6_restore_cb(skb); |
| 1510 | goto no_tcp_socket; |
| 1511 | case TCP_TW_SUCCESS: |
| 1512 | ; |
| 1513 | } |
| 1514 | goto discard_it; |
| 1515 | } |
| 1516 | |
| 1517 | static void tcp_v6_early_demux(struct sk_buff *skb) |
| 1518 | { |
| 1519 | const struct ipv6hdr *hdr; |
| 1520 | const struct tcphdr *th; |
| 1521 | struct sock *sk; |
| 1522 | |
| 1523 | if (skb->pkt_type != PACKET_HOST) |
| 1524 | return; |
| 1525 | |
| 1526 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) |
| 1527 | return; |
| 1528 | |
| 1529 | hdr = ipv6_hdr(skb); |
| 1530 | th = tcp_hdr(skb); |
| 1531 | |
| 1532 | if (th->doff < sizeof(struct tcphdr) / 4) |
| 1533 | return; |
| 1534 | |
| 1535 | /* Note : We use inet6_iif() here, not tcp_v6_iif() */ |
| 1536 | sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo, |
| 1537 | &hdr->saddr, th->source, |
| 1538 | &hdr->daddr, ntohs(th->dest), |
| 1539 | inet6_iif(skb)); |
| 1540 | if (sk) { |
| 1541 | skb->sk = sk; |
| 1542 | skb->destructor = sock_edemux; |
| 1543 | if (sk_fullsock(sk)) { |
| 1544 | struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst); |
| 1545 | |
| 1546 | if (dst) |
| 1547 | dst = dst_check(dst, inet6_sk(sk)->rx_dst_cookie); |
| 1548 | if (dst && |
| 1549 | inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) |
| 1550 | skb_dst_set_noref(skb, dst); |
| 1551 | } |
| 1552 | } |
| 1553 | } |
| 1554 | |
| 1555 | static struct timewait_sock_ops tcp6_timewait_sock_ops = { |
| 1556 | .twsk_obj_size = sizeof(struct tcp6_timewait_sock), |
| 1557 | .twsk_unique = tcp_twsk_unique, |
| 1558 | .twsk_destructor = tcp_twsk_destructor, |
| 1559 | }; |
| 1560 | |
| 1561 | static const struct inet_connection_sock_af_ops ipv6_specific = { |
| 1562 | .queue_xmit = inet6_csk_xmit, |
| 1563 | .send_check = tcp_v6_send_check, |
| 1564 | .rebuild_header = inet6_sk_rebuild_header, |
| 1565 | .sk_rx_dst_set = inet6_sk_rx_dst_set, |
| 1566 | .conn_request = tcp_v6_conn_request, |
| 1567 | .syn_recv_sock = tcp_v6_syn_recv_sock, |
| 1568 | .net_header_len = sizeof(struct ipv6hdr), |
| 1569 | .net_frag_header_len = sizeof(struct frag_hdr), |
| 1570 | .setsockopt = ipv6_setsockopt, |
| 1571 | .getsockopt = ipv6_getsockopt, |
| 1572 | .addr2sockaddr = inet6_csk_addr2sockaddr, |
| 1573 | .sockaddr_len = sizeof(struct sockaddr_in6), |
| 1574 | .bind_conflict = inet6_csk_bind_conflict, |
| 1575 | #ifdef CONFIG_COMPAT |
| 1576 | .compat_setsockopt = compat_ipv6_setsockopt, |
| 1577 | .compat_getsockopt = compat_ipv6_getsockopt, |
| 1578 | #endif |
| 1579 | .mtu_reduced = tcp_v6_mtu_reduced, |
| 1580 | }; |
| 1581 | |
| 1582 | #ifdef CONFIG_TCP_MD5SIG |
| 1583 | static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = { |
| 1584 | .md5_lookup = tcp_v6_md5_lookup, |
| 1585 | .calc_md5_hash = tcp_v6_md5_hash_skb, |
| 1586 | .md5_parse = tcp_v6_parse_md5_keys, |
| 1587 | }; |
| 1588 | #endif |
| 1589 | |
| 1590 | /* |
| 1591 | * TCP over IPv4 via INET6 API |
| 1592 | */ |
| 1593 | static const struct inet_connection_sock_af_ops ipv6_mapped = { |
| 1594 | .queue_xmit = ip_queue_xmit, |
| 1595 | .send_check = tcp_v4_send_check, |
| 1596 | .rebuild_header = inet_sk_rebuild_header, |
| 1597 | .sk_rx_dst_set = inet_sk_rx_dst_set, |
| 1598 | .conn_request = tcp_v6_conn_request, |
| 1599 | .syn_recv_sock = tcp_v6_syn_recv_sock, |
| 1600 | .net_header_len = sizeof(struct iphdr), |
| 1601 | .setsockopt = ipv6_setsockopt, |
| 1602 | .getsockopt = ipv6_getsockopt, |
| 1603 | .addr2sockaddr = inet6_csk_addr2sockaddr, |
| 1604 | .sockaddr_len = sizeof(struct sockaddr_in6), |
| 1605 | .bind_conflict = inet6_csk_bind_conflict, |
| 1606 | #ifdef CONFIG_COMPAT |
| 1607 | .compat_setsockopt = compat_ipv6_setsockopt, |
| 1608 | .compat_getsockopt = compat_ipv6_getsockopt, |
| 1609 | #endif |
| 1610 | .mtu_reduced = tcp_v4_mtu_reduced, |
| 1611 | }; |
| 1612 | |
| 1613 | #ifdef CONFIG_TCP_MD5SIG |
| 1614 | static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { |
| 1615 | .md5_lookup = tcp_v4_md5_lookup, |
| 1616 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
| 1617 | .md5_parse = tcp_v6_parse_md5_keys, |
| 1618 | }; |
| 1619 | #endif |
| 1620 | |
| 1621 | /* NOTE: A lot of things set to zero explicitly by call to |
| 1622 | * sk_alloc() so need not be done here. |
| 1623 | */ |
| 1624 | static int tcp_v6_init_sock(struct sock *sk) |
| 1625 | { |
| 1626 | struct inet_connection_sock *icsk = inet_csk(sk); |
| 1627 | |
| 1628 | tcp_init_sock(sk); |
| 1629 | |
| 1630 | icsk->icsk_af_ops = &ipv6_specific; |
| 1631 | |
| 1632 | #ifdef CONFIG_TCP_MD5SIG |
| 1633 | tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific; |
| 1634 | #endif |
| 1635 | |
| 1636 | return 0; |
| 1637 | } |
| 1638 | |
| 1639 | static void tcp_v6_destroy_sock(struct sock *sk) |
| 1640 | { |
| 1641 | tcp_v4_destroy_sock(sk); |
| 1642 | inet6_destroy_sock(sk); |
| 1643 | } |
| 1644 | |
| 1645 | #ifdef CONFIG_PROC_FS |
| 1646 | /* Proc filesystem TCPv6 sock list dumping. */ |
| 1647 | static void get_openreq6(struct seq_file *seq, |
| 1648 | const struct request_sock *req, int i) |
| 1649 | { |
| 1650 | long ttd = req->rsk_timer.expires - jiffies; |
| 1651 | const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr; |
| 1652 | const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr; |
| 1653 | |
| 1654 | if (ttd < 0) |
| 1655 | ttd = 0; |
| 1656 | |
| 1657 | seq_printf(seq, |
| 1658 | "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " |
| 1659 | "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n", |
| 1660 | i, |
| 1661 | src->s6_addr32[0], src->s6_addr32[1], |
| 1662 | src->s6_addr32[2], src->s6_addr32[3], |
| 1663 | inet_rsk(req)->ir_num, |
| 1664 | dest->s6_addr32[0], dest->s6_addr32[1], |
| 1665 | dest->s6_addr32[2], dest->s6_addr32[3], |
| 1666 | ntohs(inet_rsk(req)->ir_rmt_port), |
| 1667 | TCP_SYN_RECV, |
| 1668 | 0, 0, /* could print option size, but that is af dependent. */ |
| 1669 | 1, /* timers active (only the expire timer) */ |
| 1670 | jiffies_to_clock_t(ttd), |
| 1671 | req->num_timeout, |
| 1672 | from_kuid_munged(seq_user_ns(seq), |
| 1673 | sock_i_uid(req->rsk_listener)), |
| 1674 | 0, /* non standard timer */ |
| 1675 | 0, /* open_requests have no inode */ |
| 1676 | 0, req); |
| 1677 | } |
| 1678 | |
| 1679 | static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) |
| 1680 | { |
| 1681 | const struct in6_addr *dest, *src; |
| 1682 | __u16 destp, srcp; |
| 1683 | int timer_active; |
| 1684 | unsigned long timer_expires; |
| 1685 | const struct inet_sock *inet = inet_sk(sp); |
| 1686 | const struct tcp_sock *tp = tcp_sk(sp); |
| 1687 | const struct inet_connection_sock *icsk = inet_csk(sp); |
| 1688 | const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; |
| 1689 | |
| 1690 | dest = &sp->sk_v6_daddr; |
| 1691 | src = &sp->sk_v6_rcv_saddr; |
| 1692 | destp = ntohs(inet->inet_dport); |
| 1693 | srcp = ntohs(inet->inet_sport); |
| 1694 | |
| 1695 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { |
| 1696 | timer_active = 1; |
| 1697 | timer_expires = icsk->icsk_timeout; |
| 1698 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { |
| 1699 | timer_active = 4; |
| 1700 | timer_expires = icsk->icsk_timeout; |
| 1701 | } else if (timer_pending(&sp->sk_timer)) { |
| 1702 | timer_active = 2; |
| 1703 | timer_expires = sp->sk_timer.expires; |
| 1704 | } else { |
| 1705 | timer_active = 0; |
| 1706 | timer_expires = jiffies; |
| 1707 | } |
| 1708 | |
| 1709 | seq_printf(seq, |
| 1710 | "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " |
| 1711 | "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n", |
| 1712 | i, |
| 1713 | src->s6_addr32[0], src->s6_addr32[1], |
| 1714 | src->s6_addr32[2], src->s6_addr32[3], srcp, |
| 1715 | dest->s6_addr32[0], dest->s6_addr32[1], |
| 1716 | dest->s6_addr32[2], dest->s6_addr32[3], destp, |
| 1717 | sp->sk_state, |
| 1718 | tp->write_seq-tp->snd_una, |
| 1719 | (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq), |
| 1720 | timer_active, |
| 1721 | jiffies_delta_to_clock_t(timer_expires - jiffies), |
| 1722 | icsk->icsk_retransmits, |
| 1723 | from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)), |
| 1724 | icsk->icsk_probes_out, |
| 1725 | sock_i_ino(sp), |
| 1726 | atomic_read(&sp->sk_refcnt), sp, |
| 1727 | jiffies_to_clock_t(icsk->icsk_rto), |
| 1728 | jiffies_to_clock_t(icsk->icsk_ack.ato), |
| 1729 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
| 1730 | tp->snd_cwnd, |
| 1731 | sp->sk_state == TCP_LISTEN ? |
| 1732 | fastopenq->max_qlen : |
| 1733 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh) |
| 1734 | ); |
| 1735 | } |
| 1736 | |
| 1737 | static void get_timewait6_sock(struct seq_file *seq, |
| 1738 | struct inet_timewait_sock *tw, int i) |
| 1739 | { |
| 1740 | long delta = tw->tw_timer.expires - jiffies; |
| 1741 | const struct in6_addr *dest, *src; |
| 1742 | __u16 destp, srcp; |
| 1743 | |
| 1744 | dest = &tw->tw_v6_daddr; |
| 1745 | src = &tw->tw_v6_rcv_saddr; |
| 1746 | destp = ntohs(tw->tw_dport); |
| 1747 | srcp = ntohs(tw->tw_sport); |
| 1748 | |
| 1749 | seq_printf(seq, |
| 1750 | "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " |
| 1751 | "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n", |
| 1752 | i, |
| 1753 | src->s6_addr32[0], src->s6_addr32[1], |
| 1754 | src->s6_addr32[2], src->s6_addr32[3], srcp, |
| 1755 | dest->s6_addr32[0], dest->s6_addr32[1], |
| 1756 | dest->s6_addr32[2], dest->s6_addr32[3], destp, |
| 1757 | tw->tw_substate, 0, 0, |
| 1758 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, |
| 1759 | atomic_read(&tw->tw_refcnt), tw); |
| 1760 | } |
| 1761 | |
| 1762 | static int tcp6_seq_show(struct seq_file *seq, void *v) |
| 1763 | { |
| 1764 | struct tcp_iter_state *st; |
| 1765 | struct sock *sk = v; |
| 1766 | |
| 1767 | if (v == SEQ_START_TOKEN) { |
| 1768 | seq_puts(seq, |
| 1769 | " sl " |
| 1770 | "local_address " |
| 1771 | "remote_address " |
| 1772 | "st tx_queue rx_queue tr tm->when retrnsmt" |
| 1773 | " uid timeout inode\n"); |
| 1774 | goto out; |
| 1775 | } |
| 1776 | st = seq->private; |
| 1777 | |
| 1778 | if (sk->sk_state == TCP_TIME_WAIT) |
| 1779 | get_timewait6_sock(seq, v, st->num); |
| 1780 | else if (sk->sk_state == TCP_NEW_SYN_RECV) |
| 1781 | get_openreq6(seq, v, st->num); |
| 1782 | else |
| 1783 | get_tcp6_sock(seq, v, st->num); |
| 1784 | out: |
| 1785 | return 0; |
| 1786 | } |
| 1787 | |
| 1788 | static const struct file_operations tcp6_afinfo_seq_fops = { |
| 1789 | .owner = THIS_MODULE, |
| 1790 | .open = tcp_seq_open, |
| 1791 | .read = seq_read, |
| 1792 | .llseek = seq_lseek, |
| 1793 | .release = seq_release_net |
| 1794 | }; |
| 1795 | |
| 1796 | static struct tcp_seq_afinfo tcp6_seq_afinfo = { |
| 1797 | .name = "tcp6", |
| 1798 | .family = AF_INET6, |
| 1799 | .seq_fops = &tcp6_afinfo_seq_fops, |
| 1800 | .seq_ops = { |
| 1801 | .show = tcp6_seq_show, |
| 1802 | }, |
| 1803 | }; |
| 1804 | |
| 1805 | int __net_init tcp6_proc_init(struct net *net) |
| 1806 | { |
| 1807 | return tcp_proc_register(net, &tcp6_seq_afinfo); |
| 1808 | } |
| 1809 | |
| 1810 | void tcp6_proc_exit(struct net *net) |
| 1811 | { |
| 1812 | tcp_proc_unregister(net, &tcp6_seq_afinfo); |
| 1813 | } |
| 1814 | #endif |
| 1815 | |
| 1816 | static void tcp_v6_clear_sk(struct sock *sk, int size) |
| 1817 | { |
| 1818 | struct inet_sock *inet = inet_sk(sk); |
| 1819 | |
| 1820 | /* we do not want to clear pinet6 field, because of RCU lookups */ |
| 1821 | sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6)); |
| 1822 | |
| 1823 | size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6); |
| 1824 | memset(&inet->pinet6 + 1, 0, size); |
| 1825 | } |
| 1826 | |
| 1827 | struct proto tcpv6_prot = { |
| 1828 | .name = "TCPv6", |
| 1829 | .owner = THIS_MODULE, |
| 1830 | .close = tcp_close, |
| 1831 | .connect = tcp_v6_connect, |
| 1832 | .disconnect = tcp_disconnect, |
| 1833 | .accept = inet_csk_accept, |
| 1834 | .ioctl = tcp_ioctl, |
| 1835 | .init = tcp_v6_init_sock, |
| 1836 | .destroy = tcp_v6_destroy_sock, |
| 1837 | .shutdown = tcp_shutdown, |
| 1838 | .setsockopt = tcp_setsockopt, |
| 1839 | .getsockopt = tcp_getsockopt, |
| 1840 | .recvmsg = tcp_recvmsg, |
| 1841 | .sendmsg = tcp_sendmsg, |
| 1842 | .sendpage = tcp_sendpage, |
| 1843 | .backlog_rcv = tcp_v6_do_rcv, |
| 1844 | .release_cb = tcp_release_cb, |
| 1845 | .hash = inet_hash, |
| 1846 | .unhash = inet_unhash, |
| 1847 | .get_port = inet_csk_get_port, |
| 1848 | .enter_memory_pressure = tcp_enter_memory_pressure, |
| 1849 | .stream_memory_free = tcp_stream_memory_free, |
| 1850 | .sockets_allocated = &tcp_sockets_allocated, |
| 1851 | .memory_allocated = &tcp_memory_allocated, |
| 1852 | .memory_pressure = &tcp_memory_pressure, |
| 1853 | .orphan_count = &tcp_orphan_count, |
| 1854 | .sysctl_mem = sysctl_tcp_mem, |
| 1855 | .sysctl_wmem = sysctl_tcp_wmem, |
| 1856 | .sysctl_rmem = sysctl_tcp_rmem, |
| 1857 | .max_header = MAX_TCP_HEADER, |
| 1858 | .obj_size = sizeof(struct tcp6_sock), |
| 1859 | .slab_flags = SLAB_DESTROY_BY_RCU, |
| 1860 | .twsk_prot = &tcp6_timewait_sock_ops, |
| 1861 | .rsk_prot = &tcp6_request_sock_ops, |
| 1862 | .h.hashinfo = &tcp_hashinfo, |
| 1863 | .no_autobind = true, |
| 1864 | #ifdef CONFIG_COMPAT |
| 1865 | .compat_setsockopt = compat_tcp_setsockopt, |
| 1866 | .compat_getsockopt = compat_tcp_getsockopt, |
| 1867 | #endif |
| 1868 | #ifdef CONFIG_MEMCG_KMEM |
| 1869 | .proto_cgroup = tcp_proto_cgroup, |
| 1870 | #endif |
| 1871 | .clear_sk = tcp_v6_clear_sk, |
| 1872 | }; |
| 1873 | |
| 1874 | static const struct inet6_protocol tcpv6_protocol = { |
| 1875 | .early_demux = tcp_v6_early_demux, |
| 1876 | .handler = tcp_v6_rcv, |
| 1877 | .err_handler = tcp_v6_err, |
| 1878 | .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL, |
| 1879 | }; |
| 1880 | |
| 1881 | static struct inet_protosw tcpv6_protosw = { |
| 1882 | .type = SOCK_STREAM, |
| 1883 | .protocol = IPPROTO_TCP, |
| 1884 | .prot = &tcpv6_prot, |
| 1885 | .ops = &inet6_stream_ops, |
| 1886 | .flags = INET_PROTOSW_PERMANENT | |
| 1887 | INET_PROTOSW_ICSK, |
| 1888 | }; |
| 1889 | |
| 1890 | static int __net_init tcpv6_net_init(struct net *net) |
| 1891 | { |
| 1892 | return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6, |
| 1893 | SOCK_RAW, IPPROTO_TCP, net); |
| 1894 | } |
| 1895 | |
| 1896 | static void __net_exit tcpv6_net_exit(struct net *net) |
| 1897 | { |
| 1898 | inet_ctl_sock_destroy(net->ipv6.tcp_sk); |
| 1899 | } |
| 1900 | |
| 1901 | static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list) |
| 1902 | { |
| 1903 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6); |
| 1904 | } |
| 1905 | |
| 1906 | static struct pernet_operations tcpv6_net_ops = { |
| 1907 | .init = tcpv6_net_init, |
| 1908 | .exit = tcpv6_net_exit, |
| 1909 | .exit_batch = tcpv6_net_exit_batch, |
| 1910 | }; |
| 1911 | |
| 1912 | int __init tcpv6_init(void) |
| 1913 | { |
| 1914 | int ret; |
| 1915 | |
| 1916 | ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP); |
| 1917 | if (ret) |
| 1918 | goto out; |
| 1919 | |
| 1920 | /* register inet6 protocol */ |
| 1921 | ret = inet6_register_protosw(&tcpv6_protosw); |
| 1922 | if (ret) |
| 1923 | goto out_tcpv6_protocol; |
| 1924 | |
| 1925 | ret = register_pernet_subsys(&tcpv6_net_ops); |
| 1926 | if (ret) |
| 1927 | goto out_tcpv6_protosw; |
| 1928 | out: |
| 1929 | return ret; |
| 1930 | |
| 1931 | out_tcpv6_protosw: |
| 1932 | inet6_unregister_protosw(&tcpv6_protosw); |
| 1933 | out_tcpv6_protocol: |
| 1934 | inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); |
| 1935 | goto out; |
| 1936 | } |
| 1937 | |
| 1938 | void tcpv6_exit(void) |
| 1939 | { |
| 1940 | unregister_pernet_subsys(&tcpv6_net_ops); |
| 1941 | inet6_unregister_protosw(&tcpv6_protosw); |
| 1942 | inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP); |
| 1943 | } |